Timeevo improves time series question answering agents with self-driven tool growth
TimeEvo: Failure-Driven Self-Evolution of a Time Series Agent
Artificial IntelligenceMachine Learning
Summary
Time series agents use tools chosen beforehand to answer questions, but some tools help with some questions and hurt with others. The authors found that simply adding expert tools can reduce accuracy and that self-revised answers may change without improving results. To fix this, the authors created TimeEvo, which identifies where an agent fails, designs new tools to fix those gaps, and only accepts helpful tools. Tests show TimeEvo improves accuracy on multiple tasks and works well even when tools are grown using weaker models.
What this means in practice
- •For time series analysts: Improve the accuracy of automated analytical agents by enabling them to grow their own specialized tools based on failure patterns in time series data.
- •For data engineering teams: Deploy adaptive time series question answering agents that systematically expand and refine their toolkits to handle diverse tasks across models of varying strength.
Authors
Jie Yang, Yan Zheng, Jiarui Sun, Xiran Fan, Junpeng Wang, Liang Wang, Zelin Xu, Qinghua Liu, Zhengyu Fang, Yiwei Cai, Philip S. Yu
Abstract
Time series agents answer analytical questions by calling external tools, and which tools they carry is decided by people before the agent runs. However, we identify two failures in this setup. Human-Agent Tool Misalignment: a library of 21 expert-curated tools helps on some tasks and hurts on others, dropping anomaly accuracy under every backbone we test. Silent Harm: one round of generic self-revision changes 147 answers and breaks 56 of them, while the final score moves by less than a point. Both follow from the same gap: whether a tool helps is decided question by question at runtime, while tools are supplied in advance and judged by a single average. To address this, we propose TimeEvo, which clusters an agent's diagnosed failures into capability gaps, plans a measurement for each, synthesizes evidence-only tools that fill them, and admits the candidate library only through a paired admission gate. Experiments on ten time series QA tasks and three backbones show that TimeEvo, starting from an empty library, improves accuracy on every task and every backbone, and that a library grown on a cheap model still gains when it is installed into stronger ones. Code is available at https://github.com/Muyiiiii/TimeEvo.